Skip to content

Dataset · microsoft

One million synthetic AgentInstruct instructions

microsoft/orca-agentinstruct-1M-v1

Synthetic English instruction-response pairs for training agentic models.

What's inside

Contains approximately 1 million instruction-response pairs generated entirely by the AgentInstruct framework. Both prompts and responses are synthetic, seeded from public web text. Covers diverse skills, including text editing, in English.

What you can do with it

  • SFT for agentic LLMs
  • Training models for instruction following
  • Validating the AgentInstruct synthetic data pipeline
Tasks
Question answering
Languages
English
Size
1M–10M rows
Downloads this month
2.2K
Likes
477

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("microsoft/orca-agentinstruct-1M-v1")